Liability for damage caused by AI is the question of who compensates for harm arising from an incorrect output of a system. The widespread belief that the technology supplier bears liability does not hold up in practice – towards a customer or third party, it is generally whoever used the system in their own operations or offered it under their own brand who is liable. The risk is limited in several ways at once: human oversight of decisions affecting people, clear labelling that a communication is automated, documented testing, records of outputs, contractual arrangements governing the relationship with the supplier, and in some cases insurance as well. The legal framework in this area is evolving, so discuss the specific situation with a lawyer and do not rely on general statements.
See also: Human oversight of AI, AI system incident, AI Supply Chain.
